0.32*** 0.32 0.37*** 0.33

ef

(0.10) (0.21) (0.12) (0.24)

log_govage À0.06*** À0.06*** À0.05*** À0.05***

(0.02) (0.02) (0.01) (0.01)

0.004 0.004 0.003 0.002

log_districts

(0.02) (0.02) (0.02) (0.02)

0.002 0.04

decentralization*

ef (0.09) (0.13)

0.21** 0.21** 0.13 0.14*

Constant

(0.09) (0.08) (0.10) (0.08)

R-squared .44 .44 .45 .44

34 34 161 161

Observations

* p < .1, ** p < .05, *** p < .01; standard errors in parentheses

81

Building Party Systems in Developing Democracies

82

95% Confidence Interval

Marginal Effect of Ethnic Fractionalization on

(a)

Inflation as Decentralization Changes (XS)

.8

Marginal Effect of Ethnic Fractionalization

.6

.4

.2

0

“.2

0 1 2 3

Decentralization

Marginal Effect of Ethnic Fractionalization on

(b)

Inflation as Decentralization Changes (Pooled)

Marginal Effect of Ethnic Fractionalization

.8

.6

.4

.2

0

“.2

0 .5 1 1.5 2

Decentralization

¬gure 3.5. Marginal Effect of Ethnic Fractionalization on In¬‚ation as

Decentralization Increases

Testing the Theory 83

table 3.9. Predicted Values Table

Value of Key Independent Variables (holding all other Predicted

variables at their means) In¬‚ation Score

Table 3.3 Model 6

.09

horizontal is 0

.19

horizontal is 1

.29

horizontal is 2

.17

subrevgdp is at its means (8.25)

.19

subrevgdp is one standard deviation above the mean

Table 3.4 Model 3 (Interaction Model)

horizontal is 0 and subrevgdp is at its mean (8.25) .06

horizontal is 1 and subrevgdp is at its mean (8.25) .15

horizontal is 2 and subrevgdp is at its mean (8.25) .24

horizontal is 0 and subrevgdp is 0 .01

horizontal is 0 and subrevgdp is one standard deviation .11

above the mean

Table 3.4 Model 4

.06

decentralization is 0

.13

decentralization is 1

.20

decentralization is 2

Table 3.5 Model 5 (Interaction Model)

.08

decentralization is 0 and probability is 0

.13

decentralization is 1 and probability is 0

.18

decentralization is 2 and probability is 0

.19

decentralization is 1 and probability is 1

.42

decentralization is 2 and probability is 1

Table 3.6 Model 5 (Interaction Model)

ENPres is 2 and proximity is 1 .15

ENPres is 3 and proximity is 1 .19

decentralization is at its mean (1.36) and proximity is 1 .14

decentralization is one standard deviation (1.98) above the .22

mean and proximity is 1

Table 3.8 Model 4 (Interaction Model)

decentralization is 0 and ef is at its mean (.3) .06

decentralization is 1 and ef is at its mean (.3) .13

decentralization is 2 and ef is at its mean (.3) .20

decentralization and ef are at their means .15

decentralization it at its mean and ef is one standard .24

deviation above the mean (.53) .29

decentralization is 2 and ef is one standard deviation above

the mean (.53)

Predicted

Table 3.7 Model 7 (Interaction Model) ENPres

no_reelection is 0 and runoff is 0 2.4

no_reelection is 1 and runoff is 0 2.9

no_reelection is 1 and runoff is 1 3.5

Building Party Systems in Developing Democracies

84

3.7 conclusion

The analysis in this chapter provides substantial support for the theory I

advanced in Chapter 2. I ¬nd that aggregation is indeed the product of

two factors: (a) the size of the aggregation payoff and (b) the proba-

bility that the largest legislative party will capture that payoff. In

addition, the focus solely on the distribution of power and resources

between national and subnational actors (vertical centralization)

misses a key part of the institutional story. Horizontal centralization, or

the distribution of power within the national government, combines

with vertical centralization to affect the size of the aggregation payoff

and shape aggregation incentives. The results show that even where

there is a high degree of vertical centralization, horizontal decentrali-

zation is enough to undermine cross-district coordination. The reverse

is also true. When power is concentrated within the national govern-

ment party system, in¬‚ation can still occur if subnational actors control

substantial shares of power and resources.

In terms of the components of horizontal centralization, bicameral-

ism and reserve domains are both important, but there was little support

for the argument that party factionalism shapes aggregation through its

effect on horizontal centralization. At this stage, it is dif¬cult to judge

whether the poor showing of the factionalism hypothesis should be

ascribed to a shortcoming in the theory or whether it is the result of the

choice of a poor proxy for party factionalism. In Chapters 4 and 5, I will

revisit the factionalism hypothesis as part of the effort to explain

aggregation in Thailand.

In addition to the size of the payoff, the evidence suggests that the

probability of capturing that prize also plays an important role in

shaping aggregation incentives. In parliamentary systems, a low proba-

bility that the largest legislative party will also capture the premiership

undermines aggregation. As the chance that someone other than the

leader of the largest party will become prime minister grows, the mar-

ginal effect of a decrease in the aggregation payoff is ampli¬ed. Party

system in¬‚ation is at its highest where a low probability combines with a

small aggregation payoff.

In presidential systems, the probability of capturing executive of¬ce

is a function of the number of presidential candidates and the proximity

of presidential and legislative elections. Similar to other studies, I ¬nd

Testing the Theory 85

that proximate elections lower the number of electoral parties but only

where the effective number of presidential candidates is low. A unique

contribution of this study is to demonstrate that proximity and the

number of presidential candidates also have an effect on aggregation.

Proximity lowers party system in¬‚ation where there are relatively few

presidential candidates, and the number of presidential candidates

itself has a substantial negative impact on cross-district coordination.

The more presidential candidates there are, the more dif¬cult it is for

legislative candidates, voters, parties, and donors to identify and

coordinate around the frontrunners. The cost is poorer aggregation.

There is also evidence that the effective number of presidential candi-

dates tends to be higher where incumbents do not run for reelection.

The results of the large-N analyses then generally support the theory

of aggregation incentives laid out in Chapter 2. The models behave in

the ways we would expect given the theory and the hypotheses derived

from it. In subsequent chapters, I shift my focus from a large-N cross-

national comparison to an in-depth examination of aggregation and

nationalization in two developing democracies “ Thailand and the

Philippines. These two countries have both undergone institutional

reforms that the theory predicts should alter aggregation incentives. As

such, Thailand and the Philippines are useful natural experiments that I

can use to further test the theory. In addition, an in-depth analysis of

coordination in these two countries also allows me to move beyond the

correlations and associations established in this chapter toward iden-

tifying some of the causal mechanisms that link the theory™s explana-

tory variables to this study™s dependent variables: aggregation, and,

ultimately, the number of parties and party system nationalization.

4

Aggregation, Nationalization, and the Number

of Parties in Thailand

4.1 introduction

In the previous two chapters, I developed and tested a theory of

aggregation incentives. In the next three chapters, I use the theory to

help explain the nature of party system development in Thailand and

the Philippines. As discussed in the introduction, Thailand and the

Philippines provide interesting variations on both the dependent and

independent variables, which allow me to further investigate the

causal mechanisms lying between the explanatory variables of interest

(the size of the aggregation payoff and probability of capturing that

prize) and the outcome of interest “ the degree of aggregation. Even

though elections in each country have often produced a comparatively

large number of parties at the national level, aggregation has generally

been much better in the Philippines vis--vis Thailand. In the following

a

three chapters, I ¬rst describe the nature of intra-district and cross-

district coordination in Thailand and the Philippines, utilizing unique

datasets of district-level electoral returns in each country. Then, using

the theory described in Chapter 2, I explain why the party system in

each country looks as it does. In so doing, I answer the question of

why, given similar majoritarian electoral institutions, there has until

recently been more parties in Thailand than in the Philippines. Finally,

I utilize episodes of institutional reform in each country to conduct

comparative statics tests of the theory. The theory helps account for (a)

the dramatic fall in the number of national parties after the 1997

86

Thailand: Aggregation, Nationalization, Number of Parties 87

constitutional reforms in Thailand and (b) the demise of the two-party

system since the return of democracy in the Philippines in 1986.

In this chapter and the next, the focus is on coordination, the number

of parties, and nationalization in Thailand. In Chapter 5, I marry the

theory of aggregation incentives with the Thai case in an effort to both

deepen the theory and illuminate the nature of the Thai party system. In

this chapter, I lay the necessary groundwork for that analysis by taking

an in-depth look at the Thai party system. This is useful both because

Thailand is a country that is unfamiliar to most readers and because the

electoral system used during most of Thailand™s electoral history (the

block vote) is one that is comparatively rare (and hence unstudied).1

The remainder of the chapter proceeds as follows. I ¬rst brie¬‚y review

the history of the Thai party system and provide a basic description of its

characteristics.2 I then describe the features of the Thai electoral system in

use from 1978 to 1996 and derive a set of hypotheses regarding coordi-

nation and the number of parties at the district level.3 Thailand™s unusual

electoral system allows for an ideal comparative statics test of electoral

theories such as Duverger™s law and Cox™s M þ 1 rule. Using district-level

election data, I assess the validity of these hypotheses and explore devia-

tions from the mean across time and across regions in Thailand. This focus